The argument of this book has been narrower than the sector’s usual claims, and I want to restate it plainly at the end. Artificial intelligence is no longer limited by chips or by the price of electricity. It is limited by the queue for power, by transformers, permits and interconnection studies measured in years. Orbit is the one place where a very large amount of energy is already available and unqueued, and since 2024 the price of getting there has fallen far enough to make using it arguable. Going to orbit does not remove constraints. It substitutes a scheduling problem for a physics problem: an interconnection queue for a radiator, a transformer order book for a launch manifest. That trade is good if, and only if, the physics problem is smaller than the scheduling problem. Worked through honestly in Part V, the answer today is: not yet for a commodity buyer, already yes for a premium one and the gap between those two closes on a curve, not on a hope. That curve is the argument of this book. At today’s costs, orbital compute is a first-of-a-kind industry building singledigit units at venture rates, and it looks expensive, exactly as solar looked in 2010 and reusable launch looked in 2012. Apply the learning rates that every manufactured hardware industry has followed, add the financing maturation that every infrastructure asset class has followed, and the arithmetic crosses the commodity price somewhere between 2031 and 2034, with nothing new invented, no breakthrough required, and revenue held deliberately flat throughout. Premium buyers fund the descent. Defence, sovereign and sensor-adjacent customers pay enough today to buy the first nine doublings of manufacturing experience, and those doublings are what deliver the commodity market later. That sequence, pay a premium early, industrialise, cross over is not a workaround. It is how every infrastructure industry in history has been built. But the ideas underneath it are old, and they are not going away. Kardashev’s observation that a civilisation is bounded by the energy it can gather; Glaser’s satellite; O’Neill’s high frontier; Dyson’s insight that the waste heat is
the visible signature of thinking at scale. None of those people were wrong. They were early, by a factor of about fifty thousand dollars per kilogram. What changed is a price, and prices keep moving. What follows from it is engineering and the engineering, part by part, is what this book has been about. If the curve holds, the 2030s are the decade in which computation stops being something a civilisation does on its planet’s surface and starts being something it does in its planet’s sunlight. That is a large claim, and this book has tried to earn it the only way a claim like that can be earned: by writing down every number, showing where it came from, and marking clearly the four that could still prove it wrong.